Phantom

About

Phantom is a voice-powered AI agent that lives in your Chrome side panel. You talk to it, it talks back — and while you're having a conversation, it can see your screen, click buttons, fill forms, scroll pages, and navigate tabs on your behalf.
Powered by the Gemini Live API for real-time bidirectional audio streaming, Phantom goes beyond simple chatbots. It's an AI that can see, hear, and act inside your browser.
KEY FEATURES
• Real-time voice conversations — Talk naturally with 30+ HD voices. The AI reads your tone and responds with emotion.
• 20 browser automation tools — Phantom clicks, types, scrolls, highlights, and navigates autonomously based on your voice commands.
• Computer Use (AI Vision) — Phantom looks at your screen and clicks at exact pixel coordinates. Works on canvas elements, iframes, video players — anything visible on screen.
• Live screen vision — Streams your screen at 1fps so the AI can see what you see and react to changes in real time.
• Tab audio streaming — Phantom hears what's playing in your tab (videos, podcasts, music) and can respond to it.
• Persistent memory — Remembers you across sessions using local vector embeddings. Your name, preferences, and past conversations carry over.
• Privacy Shield — Automatically blurs passwords, credit cards, SSNs, and API keys before any screenshot reaches the AI. Your secrets never leave your device.
• 9 unique personas — Each with its own voice, pixel-art mascot, and personality. Pick a detective, pirate, wizard, or gremlin as your browser companion.
HOW IT WORKS
1. Open the Phantom side panel
2. Pick a persona
3. Tap the mic and start talking
Say things like:
- "Open YouTube and search for lo-fi music"
- "Click the sign-in button"
- "Read this page and summarize it"
- "Fill in the form with my info"
- "What's playing in this video?"
Phantom connects to the Gemini Live API through a secure Cloud Run proxy. Your voice streams as audio, the AI responds with voice + tool calls, and actions execute directly in your browser.
PRIVACY
All memory and settings are stored locally in Chrome. The Privacy Shield scans every frame before it's sent to the AI, automatically blurring sensitive content like passwords, credit card numbers, and API keys. No data is sold or transferred to third parties.
TECHNOLOGY
Built with Gemini 2.5 Flash (Live API), Google Cloud Run, Plasmo framework, React, TypeScript, and Transformers.js for local embeddings.
Open source: https://github.com/youneslaaroussi/Phantom